CN109871805A - A kind of electromagnetic signal opener recognition methods - Google Patents
A kind of electromagnetic signal opener recognition methods Download PDFInfo
- Publication number
- CN109871805A CN109871805A CN201910126528.6A CN201910126528A CN109871805A CN 109871805 A CN109871805 A CN 109871805A CN 201910126528 A CN201910126528 A CN 201910126528A CN 109871805 A CN109871805 A CN 109871805A
- Authority
- CN
- China
- Prior art keywords
- electromagnetic signal
- convolutional neural
- neural networks
- sample
- parameter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
The present invention relates to a kind of electromagnetic signal opener recognition methods, belong to signal processing technology field, solve the problems, such as that existing opener recognition methods differentiates that accuracy rate is low, performance is poor.Steps are as follows: obtaining electromagnetic signal sample set and each electromagnetic signal sample generic；The electromagnetic signal sample set is divided into training set and test set；Using training set training convolutional neural networks, the convolutional neural networks after training are assessed using test set, obtain optimal convolutional neural networks；According to optimal convolutional neural networks, construction feature parameter weibull distributed model；Opener identification is carried out to unknown electromagnetic signal using optimal convolutional neural networks, and judges the accuracy of recognition result according to the characteristic parameter weibull distributed model of construction.This method effectively increases the accuracy of electromagnetic signal opener identification.
Description
Technical field
The present invention relates to signal processing technology field more particularly to a kind of electromagnetic signal opener recognition methods.
Background technique
With science and technology fast development, electromagnetic signal identification national defense safety, wisdom traffic, in terms of have
It is widely applied and important researching value.Electromagnetic signal identification on ordinary meaning, which refers to, inputs electromagnetic signal sample by calculating
Originally the recognition result of input signal is provided with the similarity of sample in known signal library.Accordingly, there exist two classes to identify problem: 1)
Closed set (Close Set) electromagnetic signal of no rejection identifies, that is, assumes that the electromagnetic signal sample of input centainly belongs in signal library
Some individual；2) there is the opener (Open Set) of rejection to identify, i.e., whether input electromagnetic signal sample is belonged to first known
Signal library judges, and provides recognition result again on the basis of determination.In reality scene, electromagnetic identification systems face more
More is open electromagnetic environment, not only there is known signal, and there are also unknown signalings.And the opposite closed set signal identification without rejection,
The signal identification of opener can effectively distinguish unknown signaling and known signal, thus be more in line with identifying system in practical application,
More research significance.
Identify have numerous recognizers of good performance at present for closed set electromagnetic signal, such as Fisher linear discriminant,
Garbor feature decision classification is attained by higher correct recognition rata, but these algorithms are used in the performance of opener identification
Not fully up to expectations, therefore opener electromagnetic signal identification problem is got more attention.Nearest neighbor algorithm seek test sample with
Know apart from reckling between each sample of classification, by being compared with preset threshold value, to determine reception or refusal test sample
Affiliated class, but recognition effect is seriously affected by factors such as noise, disturbances.To eliminate these variations, researcher is proposed
Normalized method carries out opener using the smallest normalized cumulant and differentiates to improve differentiation accuracy rate.Although above-mentioned opener is known
Other method all reduces false acceptance rate in the case where same false rejection rate, but these methods only utilize apart from reckling or
It is the information progress opener differentiation of this dimension of confidence level the maximum, has given up between test sample and each known class sample
It include a large amount of discriminant information in the distribution of distance or confidence level in space, therefore, the accuracy rate that above-mentioned opener differentiates is lower,
Performance is unsatisfactory.
Summary of the invention
In view of above-mentioned analysis, the present invention is intended to provide a kind of electromagnetic signal opener recognition methods, to solve existing open
Set identification method differentiates the problem that accuracy rate is low, performance is poor.
The purpose of the present invention is mainly achieved through the following technical solutions:
A kind of electromagnetic signal opener recognition methods, steps are as follows:
Obtain electromagnetic signal sample set and each electromagnetic signal sample generic；
The electromagnetic signal sample set is divided into training set and test set；Utilize training set training convolutional neural networks, benefit
The convolutional neural networks after training are assessed with test set, obtain optimal convolutional neural networks；
According to optimal convolutional neural networks, construction feature parameter weibull distributed model；
Opener identification is carried out to unknown electromagnetic signal using optimal convolutional neural networks, and according to the characteristic parameter of construction
Weibull distributed model judges the accuracy of recognition result.
The present invention has the beneficial effect that: using the powerful characteristic present ability of convolutional neural networks, introducing extreme value theory, fills
Divide using the sample parameter correctly detected, construction feature parameter weibull distributed model, and according to the characteristic parameter of construction
Weibull distributed model judges the accuracy of unknown electromagnetic signal recognition result, can obtain preferable unknown sample rejection knot
Fruit.
On the basis of above scheme, the present invention has also done following improvement:
Further, described to include: using training set training convolutional neural networks
Using each electromagnetic signal sample in the training set as the input of convolutional neural networks, by corresponding electromagnetic signal
Output of the sample generic as convolutional neural networks, training convolutional neural networks.
Further, the convolutional neural networks successively include K Primary layer and full articulamentum；K >=1 and K are integer；Its
In,
Each Primary layer includes convolutional layer, ReLu layers and pond layer, and wherein convolutional layer carries out process of convolution to signal, extracts
Feature；ReLu layers of offer coefficient carry out nonlinear transformation；Pond layer compresses input feature vector figure, extracts main feature, drop
Low network query function complexity；
Full articulamentum includes active coating, classification layer, wherein active coating is used to connect all spies of k-th Primary layer output
Sign, and all features of output are classified by layer of classifying.
Further, the classification layer realizes classification feature using softmax classifier.
Beneficial effect using above-mentioned further scheme is: softmax classifier can be realized the classification of plurality of classes, tool
Preparation Method is simple, easy to accomplish, high accuracy for examination.
Further, described according to optimal convolutional neural networks, construction feature parameter weibull distributed model further wraps
Include following steps:
Take out the activation layer coefficients v in optimal convolutional neural networks1(x)、...、vc(x), wherein c is test result classification
Number；
The electromagnetic signal sample correctly classified in test process is activated into layer coefficients, is denoted as Si,j=vj(xi,j), wherein i is
The corresponding serial number of i-th of electromagnetic signal sample, j are the corresponding sample class of i-th of electromagnetic signal sample, j=1,2 ..., c；
Layer coefficients are activated according to the electromagnetic signal sample correctly classified in test process, in the mean value for calculating every a kind of sample
The heart is denoted as uj=meani(Si,j)；
Calculate the electromagnetic signal sample activation layer coefficients and respective class mean value center correctly classified in every a kind of sample away from
It is ranked up from descending, obtains activating layer coefficients collating sequence S accordinglyj(l), take preceding several items in collating sequence into
Row weibull fitting of distribution, obtains parameter distribution model, is denoted as:
ρj=(τj,κj,λj)=weibullfit (| | Sj(l)-μj||,η) (1)
Wherein, τj,κj,λjDisplacement parameter, form parameter and the scale parameter of respectively parameter weibull distribution, η are choosing
Take collating sequence Sj(l) item number carries out weibull fitting of distribution to relevant parameter using weibullfit.
Beneficial effect using above-mentioned further scheme is: the convolutional neural networks of construction are only capable of electromagnetic signal to be identified
It is identified as one of known c class signal, by introducing weibull distributed model, the statistics for making full use of weibull to be distributed
Characteristic, the inner link between analysis activation layer coefficients and weibull fitting of distribution can be realized the electromagnetism letter to unknown classification
Number identification, improve electromagnetic signal opener identification accuracy rate.
Further, described that opener identification is carried out to unknown electromagnetic signal using optimal convolutional neural networks, and according to construction
Characteristic parameter weibull distributed model judge the accuracy of recognition result, further include steps of
Unknown electromagnetic signal is handled after obtaining corresponding electromagnetic signal sample, optimal convolutional neural networks is input to, obtains
Obtain activation layer coefficients v currently1(x)、...、vc(x), maximum coefficient v is taken outm(x), wherein 1≤m≤c, and m is positive integer；
And result is exported by classification layer and obtains coefficient vm(x) corresponding class label P；Wherein, the class label P is classification layer
M-th of label in classification results；
According to characteristic parameter weibull distributed model, v is calculatedm(x) Euclidean distance pair between P class sample average center
The cumulative probability Distribution Value w answereds:
Wherein, upIt is the mean value center of P class sample, τp,κp,λpRespectively the weibull of P class sample is distributed displacement
Parameter, form parameter and scale parameter；
It sets accumulated probability and is distributed threshold value, work as wsWhen more than or equal to the threshold value, the unknown electromagnetic signal is judged to unknown letter
Number；Work as wsLess than the threshold value, the unknown electromagnetic signal is judged to known signal, and the classification of the known signal is P.
Beneficial effect using above-mentioned further scheme is: realizing the preliminary of unknown electromagnetic signal by convolutional neural networks
Identification, but recognition result is not necessarily accurate, because convolutional neural networks are only capable of unknown electromagnetic signal being identified as known class
One of；To eliminate this uncertainty, the standard of recognition result is judged according to the characteristic parameter weibull distributed model of construction
True property, if current cumulative probability Distribution Value is greater than accumulated probability and is distributed threshold value, unknown electromagnetic signal is judged to unknown signaling；Instead
Flutterring then is known signal, and signal classification is the recognition result of convolutional neural networks.This aspect can effectively promote electromagnetism letter
The accuracy of number opener identification.
Further, the accumulated probability distribution threshold value is arranged in the following manner:
Take out the activation layer coefficients V in optimal convolutional neural networks1(x)、...、Vc(x), wherein c is test result classification
Number；
It calculates in optimal convolutional neural networks, the corresponding cumulative probability Distribution Value of all electromagnetic signal samples in test set
ws' (k), k=1,2 ..., N-M, the maximum value for the cumulative probability distribution that 95% electromagnetic signal sample is met are set as tired
Count probability distribution threshold value.
Beneficial effect using above-mentioned further scheme is: making full use of and has verified that excessive class testing data are corresponding
Cumulative probability distribution accordingly to be arranged accumulated probability distribution threshold value, can preferably identify electromagnetic signal；Meanwhile it is a to avoid
Other result error bring influences, and the maximum value for the cumulative probability distribution that 95% electromagnetic signal sample is met is set as tired
Count probability distribution threshold value.
Further, it is able to reflect the characteristic parameter of electromagnetic signal attribute by choosing, respective handling is carried out to electromagnetic signal,
Obtain the electromagnetic signal sample set.
Further, Signal Range Feature is chosen as characteristic parameter, and each electromagnetic signal is sampled through A/D and digital quadrature becomes
It changes, obtains the corresponding road I signal xI(n) and the road Q signal xQ(n), it and is calculated as follows to obtain electromagnetic signal amplitude characteristic sequence sample
This A (n):
Wherein n is positive integer, n >=1；
The corresponding electromagnetic signal amplitude characteristic sequence samples of each electromagnetic signal obtained constitute electromagnetic signal sample set.
Beneficial effect using above-mentioned further scheme is: Signal Range Feature have acquisition it is convenient, can be more anti-
The advantages such as induction signal feature can choose Signal Range Feature as characteristic parameter, seek the corresponding electromagnetism letter of each electromagnetic signal
Number amplitude characteristic sequence samples constitute electromagnetic signal sample set.
Further, each electromagnetic signal sample generic need to be set in advance in the electromagnetic signal sample set.
It in the present invention, can also be combined with each other between above-mentioned each technical solution, to realize more preferred assembled schemes.This
Other feature and advantage of invention will illustrate in the following description, also, certain advantages can become from specification it is aobvious and
It is clear to, or understand through the implementation of the invention.The objectives and other advantages of the invention can by specification, claims with
And it is achieved and obtained in specifically noted content in attached drawing.
Detailed description of the invention
Attached drawing is only used for showing the purpose of specific embodiment, and is not to be construed as limiting the invention, in entire attached drawing
In, identical reference symbol indicates identical component.
Fig. 1 is the electromagnetic signal opener recognition methods flow chart in embodiment of the present invention；
Fig. 2 is the convolutional neural networks structural schematic diagram in embodiment of the present invention；
Fig. 3 is the electromagnetic signal opener recognition methods schematic diagram in embodiment of the present invention.
Specific embodiment
Specifically describing the preferred embodiment of the present invention with reference to the accompanying drawing, wherein attached drawing constitutes the application a part, and
Together with embodiments of the present invention for illustrating the principle of the present invention, it is not intended to limit the scope of the present invention.
A specific embodiment of the invention discloses a kind of electromagnetic signal opener recognition methods, flow chart such as Fig. 1 institute
Show, specifically includes the following steps:
Step S1: electromagnetic signal sample set and each electromagnetic signal sample generic are obtained；
It is able to reflect the characteristic parameter of electromagnetic signal attribute by choosing, respective handling is carried out to electromagnetic signal, obtains institute
State electromagnetic signal sample set.
Preferably due to Signal Range Feature have acquisition it is convenient, can more advantages such as reaction signal feature, this Shen
Please preferably obtained corresponding using Signal Range Feature as characteristic parameter by each electromagnetic signal through A/D sampling and the digital quadrature transformation
The road I signal xI(n) and the road Q signal xQ(n), it and is calculated as follows to obtain electromagnetic signal amplitude characteristic sequence samples A (n):
Wherein n is positive integer, n >=1；
The corresponding electromagnetic signal amplitude characteristic sequence samples of each electromagnetic signal obtained constitute electromagnetic signal sample set；
Preferably, each electromagnetic signal sample generic need to be set in advance in the electromagnetic signal sample set.
Preferably, if including altogether N number of electromagnetic signal sample in the electromagnetic signal sample set, c class sample of signal is belonged to
In certain is a kind of；
N number of electromagnetic signal sample generic in electromagnetic signal sample set is labeled, then each electromagnetic signal sample
Successively it is labeled as
Step S2: the electromagnetic signal sample set is divided into training set and test set；Utilize training set training convolutional nerve
Network assesses the convolutional neural networks after training using test set, obtains optimal convolutional neural networks；
Step S21: being divided into training set and test set for the electromagnetic signal sample set, further executes following operation:
Electromagnetic signal sample set is divided into training set and test set in certain proportion；Generally, training set sample size
It is slightly more than test set sample size: such as 6 to 4.
Wherein, M electromagnetic signal sample (M < N) is contained in training set, each electromagnetic signal sample successively marks are as follows:In order to improve the accuracy of later period network training, M sample need to be covered all
Classification.
Contain N-M electromagnetic signal sample in test set, each electromagnetic signal sample successively marks are as follows:In order to improve the accuracy of later period network test optimization, N-M sample
All categories need to be covered.
Step S22: utilizing training set training convolutional neural networks, further executes following operation:
Using each electromagnetic signal sample in the training set as the input of convolutional neural networks, by corresponding electromagnetic signal
Output of the sample generic as convolutional neural networks, training convolutional neural networks；
Preferably, the convolutional neural networks structural schematic diagram in the application is as indicated with 2: the convolutional neural networks successively wrap
Include K Primary layer and full articulamentum；K >=1 and K are integer；Wherein, each Primary layer includes convolutional layer, ReLu (Rectified
Linear unit corrects linear unit) layer and pond layer, wherein convolutional layer carries out process of convolution, extraction feature to signal；
ReLu layers of offer coefficient carry out nonlinear transformation；Pond layer compresses input feature vector figure, extracts main feature, reduces network
Computation complexity.Full articulamentum is further divided into active coating, classification layer,
Wherein, active coating is used to connect all features of k-th Primary layer output, and by all features of output by dividing
Class layer is classified；Preferably, the classification layer in the application realizes classification feature using softmax classifier.
Step S23: the convolutional neural networks after training are assessed using test set, obtain optimal convolutional Neural net
Network；
It, will be corresponding using each electromagnetic signal sample in the test set as the input of the convolutional neural networks after training
Output of the electromagnetic signal sample generic as convolutional neural networks after training, the property of the convolutional neural networks after assessment training
Can, and according to test accuracy rate regularized learning algorithm rate (general accuracy rate will reach 95% or more)；When accuracy rate reaches given threshold
When, algorithm training stops iteration and updates, and obtains optimal convolutional neural networks.
In actual electromagnetic signal opener identification process, in addition to the electromagnetic signal for including above-mentioned known class, it is also possible to wrap
The electromagnetic signal for including unknown classification can not identify the electromagnetic signal of unknown classification using above-mentioned convolutional neural networks,
Lead to the reduction of electromagnetic signal opener recognition accuracy.The application can not be to unknown class by introducing the solution of weibull distributed model
The problem of other electromagnetic signal is identified.
Step S3: according to optimal convolutional neural networks, construction feature parameter weibull distributed model:
Step S31: the activation layer coefficients v in optimal convolutional neural networks is taken out1(x)、...、vc(x), wherein c is test
As a result classification number；
Step S32: the electromagnetic signal sample correctly classified in test process is activated into layer coefficients, is denoted as Si,j=vj
(xi,j), wherein i is the corresponding serial number of i-th of electromagnetic signal sample, and j is the corresponding sample class of i-th of electromagnetic signal sample, j
=1,2 ..., c；
Step S33: activating layer coefficients according to the electromagnetic signal sample correctly classified in test process, calculates every a kind of sample
Mean value center, be denoted as uj=meani(Si,j)；
Step S34: the electromagnetic signal sample activation layer coefficients and respective class mean value correctly classified in every a kind of sample are calculated
The distance at center is descending to be ranked up, and obtains activating layer coefficients collating sequence S accordinglyj(l), before taking in collating sequence
Several progress weibull fittings of distribution, obtain parameter distribution model, are denoted as:
ρj=(τj,κj,λj)=weibullfit (| | Sj(l)-μj||,η) (1)
Wherein, τj,κj,λjDisplacement parameter, form parameter and the scale parameter of respectively parameter weibull distribution, η are choosing
Take collating sequence Sj(l) item number carries out weibull fitting of distribution to relevant parameter using weibullfit.
Step S4: opener identification is carried out to unknown electromagnetic signal using optimal convolutional neural networks, and according to the spy of construction
Sign parameter weibull distributed model judges the accuracy of recognition result.
Step S41: unknown electromagnetic signal is handled after obtaining corresponding electromagnetic signal sample, optimal convolutional Neural is input to
Network obtains current activation layer coefficients v1(x)、...、vc(x), maximum coefficient v is taken outm(x), wherein 1≤m≤c, and m is
Positive integer；And result is exported by classification layer and obtains coefficient vm(x) corresponding class label P；Wherein, the class label P is
M-th of label in the classification results of softmax classifier；
Step S42: according to characteristic parameter weibull distributed model, v is calculatedm(x) Europe between P class sample average center
Formula is apart from corresponding cumulative probability Distribution Value ws:
Wherein, upIt is the mean value center of P class sample, τp,κp,λpRespectively the weibull of P class sample is distributed displacement
Parameter, form parameter and scale parameter；
Step S43: setting accumulated probability is distributed threshold value, works as wsWhen more than or equal to the threshold value, the unknown electromagnetic signal is sentenced
For unknown signaling；Work as wsLess than the threshold value, the unknown electromagnetic signal is judged to known signal, and the classification of the known signal is P.
Preferably, accumulated probability distribution threshold value is arranged in the following manner:
Take out the activation layer coefficients V in optimal convolutional neural networks1(x)、...、Vc(x), wherein c is test result classification
Number；
It calculates in optimal convolutional neural networks, the corresponding cumulative probability Distribution Value of all electromagnetic signal samples in test set
ws' (k), k=1,2 ..., N-M, the maximum value for the cumulative probability distribution that 95% electromagnetic signal sample is met are set as tired
Count probability distribution threshold value.
Fig. 3 is the electromagnetic signal opener recognition methods schematic diagram in embodiment of the present invention, with above-mentioned opener identification process
It is corresponding.
A kind of electromagnetic signal opener recognition methods proposed by the present invention utilizes the powerful characteristic present energy of convolutional neural networks
Power introduces extreme value theory, makes full use of and correctly detects sample parameter, the extreme value of construction feature parameter and class centre distance
Weibull distribution calculates cumulative probability Distribution Value simultaneously according to corresponding weibull distribution situation to the active coating feature of test sample
By threshold decision signal whether it is known that preferable unknown sample rejection result can be obtained.
It will be understood by those skilled in the art that realizing all or part of the process of above-described embodiment method, meter can be passed through
Calculation machine program instruction relevant hardware is completed, and the program can be stored in computer readable storage medium.Wherein, described
Computer readable storage medium is disk, CD, read-only memory or random access memory etc..
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by anyone skilled in the art,
It should be covered by the protection scope of the present invention.
Claims (10)
1. a kind of electromagnetic signal opener recognition methods, which is characterized in that steps are as follows:
Obtain electromagnetic signal sample set and each electromagnetic signal sample generic；
The electromagnetic signal sample set is divided into training set and test set；Using training set training convolutional neural networks, survey is utilized
Examination collection assesses the convolutional neural networks after training, obtains optimal convolutional neural networks；
According to optimal convolutional neural networks, construction feature parameter weibull distributed model；
Opener identification is carried out to unknown electromagnetic signal using optimal convolutional neural networks, and according to the characteristic parameter of construction
Weibull distributed model judges the accuracy of recognition result.
2. recognition methods according to claim 1, which is characterized in that described to utilize training set training convolutional neural networks packet
It includes:
Using each electromagnetic signal sample in the training set as the input of convolutional neural networks, by corresponding electromagnetic signal sample
Output of the generic as convolutional neural networks, training convolutional neural networks.
3. recognition methods according to claim 2, which is characterized in that the convolutional neural networks successively include K basic
Layer and full articulamentum, K >=1 and K are integer；Wherein,
Each Primary layer includes convolutional layer, ReLu layers and pond layer, and the convolutional layer carries out process of convolution to sample signal, extracts
Feature；ReLu layers of offer coefficient carry out nonlinear transformation；Pond layer compresses input feature vector figure, extracts main feature；
Full articulamentum includes active coating, classification layer, wherein and active coating is used to connect all features of k-th Primary layer output, and
All features of output are classified by layer of classifying.
4. recognition methods according to claim 3, which is characterized in that the classification layer is realized using softmax classifier
Classification feature.
5. recognition methods according to claim 4, which is characterized in that described according to optimal convolutional neural networks, construction is special
Parameter weibull distributed model is levied, is further included steps of
Take out the activation layer coefficients v in optimal convolutional neural networks1(x)、...、vc(x), wherein c is test result classification number；
The electromagnetic signal sample correctly classified in test process is activated into layer coefficients, is denoted as Si,j=vj(xi,j), wherein i is i-th
The corresponding serial number of electromagnetic signal sample, j are the corresponding sample class of i-th of electromagnetic signal sample, j=1,2 ..., c；
Layer coefficients are activated according to the electromagnetic signal sample correctly classified in test process, calculate the mean value center of every a kind of sample,
It is denoted as uj=meani(Si,j)；
Calculate the electromagnetic signal sample activation layer coefficients correctly classified in every a kind of sample at a distance from respective class mean value center by
Arrive greatly it is small be ranked up, obtain activating layer coefficients collating sequence S accordinglyj(l), the preceding several progress in collating sequence are taken
Weibull fitting of distribution obtains parameter distribution model, is denoted as:
ρj=(τj,κj,λj)=weibullfit (| | Sj(l)-μj||,η) (1)
Wherein, τj,κj,λjDisplacement parameter, form parameter and the scale parameter of respectively parameter weibull distribution, η are the row of selection
Sequence sequence Sj(l) item number carries out weibull fitting of distribution to relevant parameter using weibullfit.
6. recognition methods according to any one of claims 1-5, which is characterized in that described to utilize optimal convolutional Neural net
Network carries out opener identification to unknown electromagnetic signal, and judges recognition result according to the characteristic parameter weibull distributed model of construction
Accuracy, further include steps of
Unknown electromagnetic signal is handled after obtaining corresponding electromagnetic signal sample, optimal convolutional neural networks is input to, is worked as
Preceding activation layer coefficients v1(x)、...、vc(x), maximum coefficient v is taken outm(x), wherein 1≤m≤c, and m is positive integer；And lead to
It crosses classification layer output result and obtains coefficient vm(x) corresponding class label P；Wherein, the class label P is the classification of classification layer
As a result m-th of label in；
According to characteristic parameter weibull distributed model, v is calculatedm(x) Euclidean distance is corresponding tired between P class sample average center
Product probability distribution value ws:
Wherein, upIt is the mean value center of P class sample, τp,κp,λpRespectively P class sample weibull distribution displacement parameter,
Form parameter and scale parameter；
It sets accumulated probability and is distributed threshold value, work as wsWhen more than or equal to the threshold value, the unknown electromagnetic signal is judged to unknown signaling；When
wsLess than the threshold value, the unknown electromagnetic signal is judged to known signal, and the classification of the known signal is P.
7. recognition methods according to claim 6, which is characterized in that the accumulated probability distribution threshold value is in the following manner
Setting:
Take out the activation layer coefficients V in optimal convolutional neural networks1(x)、...、Vc(x), wherein c is test result classification number；
It calculates in optimal convolutional neural networks, the corresponding cumulative probability Distribution Value w of all electromagnetic signal samples in test sets′
(k), k=1,2 ..., N-M, the maximum value for the cumulative probability distribution that 95% electromagnetic signal sample is met are set as accumulative general
Rate is distributed threshold value.
8. recognition methods according to claim 1, which is characterized in that be able to reflect the spy of electromagnetic signal attribute by choosing
Parameter is levied, respective handling is carried out to electromagnetic signal, obtains the electromagnetic signal sample set.
9. recognition methods according to claim 8, which is characterized in that selection Signal Range Feature, will as characteristic parameter
Each electromagnetic signal is sampled through A/D and the digital quadrature transformation, obtains the corresponding road I signal xI(n) and the road Q signal xQ(n), it and presses
Electromagnetic signal amplitude characteristic sequence samples A (n) is calculated in formula:
Wherein n is positive integer, n >=1；
The corresponding electromagnetic signal amplitude characteristic sequence samples of each electromagnetic signal obtained constitute electromagnetic signal sample set.
10. the recognition methods according to any one of claim 7-9, which is characterized in that in the electromagnetic signal sample set
Each electromagnetic signal sample generic need to be set in advance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910126528.6A CN109871805B (en) | 2019-02-20 | 2019-02-20 | Electromagnetic signal open set identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910126528.6A CN109871805B (en) | 2019-02-20 | 2019-02-20 | Electromagnetic signal open set identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109871805A true CN109871805A (en) | 2019-06-11 |
| CN109871805B CN109871805B (en) | 2020-10-27 |
Family
ID=66918956
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910126528.6A Active CN109871805B (en) | 2019-02-20 | 2019-02-20 | Electromagnetic signal open set identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109871805B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112163636A (en) * | 2020-10-15 | 2021-01-01 | 电子科技大学 | Unknown pattern recognition method of electromagnetic signal radiation source based on twin neural network |
| CN113705446A (en) * | 2021-08-27 | 2021-11-26 | 电子科技大学 | Open set identification method for individual radiation source |
| CN116578913A (en) * | 2023-03-31 | 2023-08-11 | 中国人民解放军陆军工程大学 | Reliable unmanned aerial vehicle detection and recognition method oriented to complex electromagnetic environment |
| CN116628619A (en) * | 2023-07-26 | 2023-08-22 | 西南交通大学 | Unknown abnormal electrical phenomenon identification method based on vehicle network coupling |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102520697A (en) * | 2011-12-16 | 2012-06-27 | 西安建筑科技大学 | Onsite information preprocessing method of remote cooperative diagnosis |
| US20140114609A1 (en) * | 2012-10-23 | 2014-04-24 | Hewlett-Packard Development Company, L.P. | Adaptive analysis of signals |
| CN104316277A (en) * | 2014-09-26 | 2015-01-28 | 西安交通大学 | Acoustic detection and blind signal separation-based air tightness monitoring method and apparatus |
| CN105654037A (en) * | 2015-12-21 | 2016-06-08 | 浙江大学 | Myoelectric signal gesture recognition method based on depth learning and feature images |
| CN106980367A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of gesture identification method based on myoelectricity topographic map |
| CN107908299A (en) * | 2017-11-16 | 2018-04-13 | 青岛罗博数码科技有限公司 | A kind of processing method of electromagnetic signal, apparatus and system |
| CN108520199A (en) * | 2018-03-04 | 2018-09-11 | 天津大学 | Based on radar image and the human action opener recognition methods for generating confrontation model |
| CN108846412A (en) * | 2018-05-08 | 2018-11-20 | 复旦大学 | A kind of method of extensive zero sample learning |
| CN109086700A (en) * | 2018-07-20 | 2018-12-25 | 杭州电子科技大学 | Radar range profile's target identification method based on depth convolutional neural networks |
| CN109342909A (en) * | 2018-12-14 | 2019-02-15 | 中国测试技术研究院电子研究所 | A kind of cable accessory Partial Discharge Pattern Recognition Method based on SLLE |
-
2019
- 2019-02-20 CN CN201910126528.6A patent/CN109871805B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102520697A (en) * | 2011-12-16 | 2012-06-27 | 西安建筑科技大学 | Onsite information preprocessing method of remote cooperative diagnosis |
| US20140114609A1 (en) * | 2012-10-23 | 2014-04-24 | Hewlett-Packard Development Company, L.P. | Adaptive analysis of signals |
| CN104316277A (en) * | 2014-09-26 | 2015-01-28 | 西安交通大学 | Acoustic detection and blind signal separation-based air tightness monitoring method and apparatus |
| CN105654037A (en) * | 2015-12-21 | 2016-06-08 | 浙江大学 | Myoelectric signal gesture recognition method based on depth learning and feature images |
| CN106980367A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of gesture identification method based on myoelectricity topographic map |
| CN107908299A (en) * | 2017-11-16 | 2018-04-13 | 青岛罗博数码科技有限公司 | A kind of processing method of electromagnetic signal, apparatus and system |
| CN108520199A (en) * | 2018-03-04 | 2018-09-11 | 天津大学 | Based on radar image and the human action opener recognition methods for generating confrontation model |
| CN108846412A (en) * | 2018-05-08 | 2018-11-20 | 复旦大学 | A kind of method of extensive zero sample learning |
| CN109086700A (en) * | 2018-07-20 | 2018-12-25 | 杭州电子科技大学 | Radar range profile's target identification method based on depth convolutional neural networks |
| CN109342909A (en) * | 2018-12-14 | 2019-02-15 | 中国测试技术研究院电子研究所 | A kind of cable accessory Partial Discharge Pattern Recognition Method based on SLLE |
Non-Patent Citations (4)
| Title |
|---|
| WALTER J.SCHEIRER 等: "Probability Models for Open Set Recognition", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| YUAN LIANG 等: "A Robust Voice Activity Detector Based on Weibull and Gaussian Mixture Distribution", 《2010 2ND INTERNATIONAL CONFERENCE ON SIGNAL PROCESSING SYSTEMS (ICSPS)》 * |
| 尹良 等: "基于机器学习的无线电信号识别", 《2017年全国无线及移动通信学术大会论文集》 * |
| 骆振兴 等: "通信辐射源个体开集识别中的二分类SVDD算法", 《通信对抗》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112163636A (en) * | 2020-10-15 | 2021-01-01 | 电子科技大学 | Unknown pattern recognition method of electromagnetic signal radiation source based on twin neural network |
| CN112163636B (en) * | 2020-10-15 | 2023-09-26 | 电子科技大学 | Unknown mode identification method of electromagnetic signal radiation source based on twin neural network |
| CN113705446A (en) * | 2021-08-27 | 2021-11-26 | 电子科技大学 | Open set identification method for individual radiation source |
| CN113705446B (en) * | 2021-08-27 | 2023-04-07 | 电子科技大学 | Open set identification method for individual radiation source |
| CN116578913A (en) * | 2023-03-31 | 2023-08-11 | 中国人民解放军陆军工程大学 | Reliable unmanned aerial vehicle detection and recognition method oriented to complex electromagnetic environment |
| CN116628619A (en) * | 2023-07-26 | 2023-08-22 | 西南交通大学 | Unknown abnormal electrical phenomenon identification method based on vehicle network coupling |
| CN116628619B (en) * | 2023-07-26 | 2023-10-20 | 西南交通大学 | Unknown abnormal electrical phenomenon identification method based on vehicle network coupling |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109871805B (en) | 2020-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109871805A (en) | A kind of electromagnetic signal opener recognition methods | |
| CN113378632B (en) | Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method | |
| CN107506702B (en) | Multi-angle-based face recognition model training and testing system and method | |
| CN109902740B (en) | Re-learning industrial control intrusion detection method based on multi-algorithm fusion parallelism | |
| CN109783879B (en) | Radar radiation source signal identification efficiency evaluation method and system | |
| CN107579846B (en) | Cloud computing fault data detection method and system | |
| CN112766229A (en) | Human face point cloud image intelligent identification system and method based on attention mechanism | |
| CN115811440B (en) | Real-time flow detection method based on network situation awareness | |
| CN115659254A (en) | Power quality disturbance analysis method for power distribution network with bimodal feature fusion | |
| Gong et al. | Multi-task based deep learning approach for open-set wireless signal identification in ISM band | |
| CN111310719B (en) | Unknown radiation source individual identification and detection method | |
| CN111797177A (en) | Financial time sequence classification method for abnormal financial account detection and application | |
| CN116633601A (en) | Detection method based on network traffic situation awareness | |
| CN113541834B (en) | Abnormal signal semi-supervised classification method and system and data processing terminal | |
| CN113987910A (en) | Method and device for identifying load of residents by coupling neural network and dynamic time planning | |
| CN117131436A (en) | Radiation source individual identification method oriented to open environment | |
| Na et al. | Fake data injection attack detection in AMI system using a hybrid method | |
| CN112418623A (en) | Anti-electricity-stealing identification method based on bidirectional long-time and short-time memory network and sliding window input | |
| CN110348323B (en) | Wearable device gesture recognition method based on neural network optimization | |
| CN104077608A (en) | Behavior recognition method based on sparsely coded slow characteristic functions | |
| Yin et al. | WiLAR: a location-adapted action recognition system based on WiFi | |
| Lin et al. | Features fusion based automatic modulation classification using convolutional neural network | |
| CN115664908B (en) | Cross-layer intelligent countermeasure method, system, medium and equipment for communication signal modulation recognition | |
| Jian et al. | 2-layer parallel SVM network based on aggregated local descriptors for fingerprint liveness detection | |
| CN112014821A (en) | Unknown vehicle target identification method based on radar broadband characteristics |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |